2011-06-13.log

--- Log opened Mon Jun 13 00:00:07 2011
-!- augur [~augur@208.58.6.161] has joined ##hplusroadmap00:01
-!- alystair [alystair@24-246-14-18.cable.teksavvy.com] has quit [Ping timeout: 240 seconds]02:12
-!- augur [~augur@208.58.6.161] has quit [Read error: Connection reset by peer]02:26
-!- augur [~augur@208.58.6.161] has joined ##hplusroadmap02:26
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has quit [Read error: No route to host]02:35
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has joined ##hplusroadmap02:35
dbolsercan anyone point me at that video hosting site with good subtitle editing functionality?02:38
dbolserI was linked to a vid on that site in here about 6 months ago02:38
Utopiahhttp://www.universalsubtitles.org ?02:39
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has quit [Read error: Connection reset by peer]02:40
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has joined ##hplusroadmap02:41
dbolserUtopiah: your fast!02:46
dbolserthanks02:47
dbolserlooks right, which is all I need :-D02:48
-!- archels [~neuralnet@541ED8B0.cm-5-7d.dynamic.ziggo.nl] has quit [Ping timeout: 240 seconds]02:53
-!- archels [~neuralnet@541ED8B0.cm-5-7d.dynamic.ziggo.nl] has joined ##hplusroadmap03:05
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has quit [Read error: Connection reset by peer]03:20
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has joined ##hplusroadmap03:21
-!- lumos [~lumos@92.20.112.33] has joined ##hplusroadmap03:52
-!- klafka [~textual@cpe-69-205-70-55.rochester.res.rr.com] has joined ##hplusroadmap04:14
-!- uniqanomaly [~ua@dynamic-78-8-90-161.ssp.dialog.net.pl] has joined ##hplusroadmap04:24
-!- lumos [~lumos@92.20.112.33] has quit [Ping timeout: 240 seconds]04:34
-!- lumos [~lumos@92.20.112.33] has joined ##hplusroadmap04:43
-!- klafka [~textual@cpe-69-205-70-55.rochester.res.rr.com] has quit [Quit: Computer has gone to sleep.]06:12
-!- QuantumG [~qg@rtfm.insomnia.org] has quit [Ping timeout: 240 seconds]06:34
-!- eridu [~eridu@gateway/tor-sasl/eridu] has joined ##hplusroadmap09:17
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has quit [Ping timeout: 252 seconds]10:01
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has joined ##hplusroadmap10:05
-!- uniqanomaly [~ua@dynamic-78-8-90-161.ssp.dialog.net.pl] has quit [Quit: uniqanomaly]12:19
-!- eudoxia [~eudoxia@r190-133-10-21.dialup.adsl.anteldata.net.uy] has joined ##hplusroadmap12:46
-!- eudoxia [~eudoxia@r190-133-10-21.dialup.adsl.anteldata.net.uy] has quit [Quit: Leaving]12:58
kanzuresenate.gov too? http://lulzsecurity.com/releases/senate.gov.txt13:10
-!- archels [~neuralnet@541ED8B0.cm-5-7d.dynamic.ziggo.nl] has quit [Ping timeout: 240 seconds]14:34
-!- archels [~neuralnet@541ED8B0.cm-5-7d.dynamic.ziggo.nl] has joined ##hplusroadmap14:45
-!- nsh_ [~nsh@cpc3-broo4-0-0-cust997.14-2.cable.virginmedia.com] has joined ##hplusroadmap14:47
augurkanzure: ping14:52
-!- QuantumG [~qg@rtfm.insomnia.org] has joined ##hplusroadmap14:54
nsh_what's new?15:03
-!- lumos [~lumos@92.20.112.33] has quit [Ping timeout: 255 seconds]15:11
kanzureaugur: pong15:15
kanzurensh_: pong15:16
augurkanzure: is there a working group here for AI?15:16
kanzurehere? what?15:17
kanzurei think that ai is possible and that everyone working on it basically sucks, except for the whole brain emulation folks who might or might not be making progress15:18
* nsh_ shrugs15:19
nsh_i don't believe in artifice :)15:20
kanzurensh_: i've been geeking out about http://diyhpl.us/~bryan/papers2/neuro/ultrasound/15:20
nsh_kanzure, what are you imagining to do with ultrasound?15:21
nsh_general biochemical tinkering?15:21
kanzuretranscranial ultrasound for stimulation and/or brain melting15:21
nsh_mmm15:22
nsh_will look into this15:30
nsh_bbl15:30
-!- nsh_ [~nsh@cpc3-broo4-0-0-cust997.14-2.cable.virginmedia.com] has quit [Ping timeout: 276 seconds]15:34
-!- PixelScum [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has joined ##hplusroadmap15:53
-!- Drakkar [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has quit [Ping timeout: 252 seconds]15:56
-!- eridu [~eridu@gateway/tor-sasl/eridu] has quit [Ping timeout: 250 seconds]16:02
-!- klafka [~textual@cpe-69-205-70-55.rochester.res.rr.com] has joined ##hplusroadmap16:11
-!- Avarice [~nsh@cpc1-cowc1-0-0-cust844.14-2.cable.virginmedia.com] has joined ##hplusroadmap16:16
-!- Avarice [~nsh@cpc1-cowc1-0-0-cust844.14-2.cable.virginmedia.com] has quit [Client Quit]16:19
-!- Avarice [~nsh@cpc1-cowc1-0-0-cust844.14-2.cable.virginmedia.com] has joined ##hplusroadmap16:20
-!- Avarice is now known as nsh_16:20
-!- nsh_ [~nsh@cpc1-cowc1-0-0-cust844.14-2.cable.virginmedia.com] has quit [Changing host]16:22
-!- nsh_ [~nsh@wikipedia/nsh] has joined ##hplusroadmap16:22
-!- nsh_ is now known as nsh16:22
-!- alystair [alystair@24-246-14-18.cable.teksavvy.com] has joined ##hplusroadmap16:27
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has quit [Read error: Connection reset by peer]16:44
jrayhawk"< aristid> instead they should grant tax exemptions." huh, that's not a bad idea17:10
kanzureseems marginally better than my "immediate nullification of all patents" plan17:12
QuantumGyou just want it both ways.  Don't like the restriction of patents, don't respect the necessity for trade secrets without patents.17:18
kanzureevery product should have its information/recipe available17:19
QuantumGthere ya go.17:22
-!- eridu [~eridu@gateway/tor-sasl/eridu] has joined ##hplusroadmap17:24
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has joined ##hplusroadmap17:41
kanzureemokit w/ websockets http://makemyactionschainreactions.net/eeg/17:46
-!- PixelScum [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has quit [Read error: Connection reset by peer]18:39
-!- PixelScum [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has joined ##hplusroadmap18:39
fennintellectual property tax makes a lot of sense18:43
fennif we're going to go down the IP road18:43
fennor secrecy tax, whatever18:44
kanzurehaha if all your other property can be taxed.. why not IP :P19:00
kanzurebut unfortunately this will be misconstrued as something to tack on, not as an alternative to limited monopolies19:00
kanzure(tax incentives vs. intellectual property tax)19:00
augurkanzure: so no. there is no AIWG here. shame!19:04
kanzureaiwg?19:04
kanzurewe have people that get excited about whole brain emulation,19:04
kanzurei think that's close enough.19:04
augurhrmph.19:05
kanzurewhat would you want19:08
-!- PixelScum [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has quit [Read error: Connection reset by peer]19:11
-!- PixelScum [~PixelScum@ip98-177-175-88.ph.ph.cox.net] has joined ##hplusroadmap19:11
kanzureis stross' "rule 34" really about rule 34?19:34
-!- eridu [~eridu@gateway/tor-sasl/eridu] has quit [Ping timeout: 250 seconds]19:39
nshaskance19:41
kanzureis that a dance style19:42
nshit's a rare spice, depreciated in modern cuisine19:43
-!- mrtrousers [5339b5dd@gateway/web/freenode/ip.83.57.181.221] has joined ##hplusroadmap20:12
-!- jennicide [jen@173-29-174-114.client.mchsi.com] has quit [Read error: Connection reset by peer]20:15
-!- jennicide [~jen@173-29-174-114.client.mchsi.com] has joined ##hplusroadmap20:15
-!- AJollyLife [~quassel@unaffiliated/ajollylife] has quit [Ping timeout: 258 seconds]20:16
foucistshe glanced at his penis in askance20:20
kanzurefoucist: RAGE20:21
kanzurehttp://heybryan.org/shots/2011-06-13-2249-arenasnow-typeracer.png20:21
kanzure<-- raped :(20:22
foucistjust an example of proper usage20:25
foucistof the word20:25
foucistin a sentence20:25
foucistheh20:25
kanzure?20:25
foucistkanzure: someone beat you? about time20:25
foucist(askance)20:25
-!- jenzebubble [jen@173-29-174-114.client.mchsi.com] has joined ##hplusroadmap20:35
-!- jennicide [~jen@173-29-174-114.client.mchsi.com] has quit [Ping timeout: 260 seconds]20:38
fenn"The other element is found in the stenotype, that somewhat disconcerting device encountered usually at public meetings. A girl strokes its keys languidly and looks about the room and sometimes at the speaker with a disquieting gaze. Combine these two elements, let the Vocoder run the stenotype, and the result is a machine which types when talked to."20:40
fenn197 wpm, is that with a querty keyboard?20:41
kanzureyes20:44
kanzurethe dude has arms of steel20:44
kanzurehttp://www.youtube.com/watch?v=uf0hrleDE-E&NR=120:44
fennthat the input box clears every word makes comparison difficult20:49
* kanzure is facebook friends with sean :(20:51
kanzureif you can't beat him.. join him20:51
fenntypaholics anonymous20:52
kanzure"But my true obsession in high school was being statistician of the Mensa Scrabble-by-Mail SIG, where I was also a top ten player before I stopped playing upon entering college"20:53
kanzurewhat20:53
fenndo intelligent people use bigger words?20:53
fenncalculemus!20:53
kanzure'mensa scrabble-by-mail SIG'20:54
kanzurei can't stop laughing :(20:54
fenni was reading julian assange's blog earlier today20:54
fennhe has this whole long diatribe about how intelligent people are socially outcast because they can't communicate with normals20:54
fennreferring specifically to 150+ iq20:55
kanzurebah my wpm is higher than that iq!20:55
kanzure(i don't know what this means)20:55
fenni dont think i'd have the patience necessary for x-by-mail games20:55
kanzurehey does 'units' handle iq20:56
fennno20:57
fenniq is based on percentile relative to the population at large20:57
fennthe average is always exactly 10020:57
* nsh mumbles something about relative-state20:58
fennunits doesn't do things that change arbitrarily, except for some currency conversions (which i think i s a bad idea anyway)20:58
fenn"almost all IQ tests adhere to the assignment of 15 IQ points to each standard deviation"21:00
fennit would be much better if they just said the mental age, me thinks21:01
kanzureso our friend the 'smartest man in the world' is, what, 12 standard deviations out?21:01
kanzureunlikely..21:01
fennthe test breaks down at high end because there arent enough to draw a baseline from when designing the test21:02
fennif you get 100 out of 100 questions right, what's your score?21:03
fennhm, this is a long wikipedia entry21:04
ybitaiwg?21:04
ybitaugur^21:04
augurO_O21:06
ybitdon't give me that face21:06
augurybit: AI working group21:06
ybitexplain yourself :)21:06
ybitokay21:06
fenn"The accepted best measure of g is Raven's Progressive Matrices which is a test of visual reasoning."21:08
fennis it just me or does that statement not make any sense21:08
fenng is supposed to be "general" not "visual"21:08
fenni guess it's accepted because it's culturally neutral21:09
augurybit: i want to work on a project to design weak AGI around more formal techniques, but atypical ones21:09
augurso i was curious if there was a wg here i could poke at21:09
ybit#opencog21:10
fennaugur: i want to pooh pooh your idea but you haven't told us enough21:10
augurfenn: im not aiming for anything strong, just something fairly agent-ish that can engage relatively fluidly in natural language.21:11
fennbut seriously i dont think any good will come of a system that does not use advanced statistics and massively parallel computation21:11
augurthe intention isnt for anything major, fenn.21:11
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has quit [Ping timeout: 260 seconds]21:11
auguri mean, honestly, i dont think that statistics will help, and im not convince that parallelism is truly necessary, but21:12
augurthats beside the point21:12
fenn... One must think "divergently" of many possible solutions to a given, often ill-defined, problem; the relative scarcity or abundance of various resources, present and projected individual or group needs...21:12
augursure thats fine. it doesnt necessitate parallelism tho21:13
fennwhat is the intention then?21:13
auguri mean, im all for parallelism. its computationally quite nice, if you can get it to work21:13
fennmegahal engages fluidly in natural language... it just doesn't say anything sensical most of the time21:14
augurbut its not magically more powerful. its equivalent to serial processing, so i see the parallelism issue as a distraction.21:14
augurbrainlike "parallelism" on the other hand is a different issue21:14
augurive never heard of megahal. ill check it out.21:14
fennit's only equivalent to serial processing if the separate threads don't interact at all21:14
auguroh yes, one of these things.21:14
fennmarkov model irc bot21:15
augurfenn: no, serial machines can simulate any parallel machine, so they're completely equivalent.21:15
augurparallelism is just useful for certain problem structures.21:15
augurfurther, interaction of threads is a concurrency issue not a parallelism issue. tho the two are often conflated, they're quite distinct.21:16
fennand happens to be the only progress in computers at the moment21:16
augurhttp://existentialtype.wordpress.com/2011/03/17/parallelism-is-not-concurrency/21:16
augurthe only progress how21:16
foucisttopological computation is the future21:17
foucistfixes all the problems w/ parallelism21:17
fennthere hasn't been significant improvement in clock speeds for years21:17
foucistfor certain classes of problems anyways21:17
augurfoucist: lazy/non-strict computation > all!21:17
augurhaskell will dominate!21:17
augurfenn: oh thats true, yes. i dont deny that parallelism is a huge thing in computation right now.21:18
-!- klafka [~textual@cpe-69-205-70-55.rochester.res.rr.com] has quit [Quit: Computer has gone to sleep.]21:18
augurbut thats a concern that should come /after/ the design of a correct program, not before. get the program to run correctly, then get it to run fast.21:19
auguralso, fenn, graphene processors promise to bump clockspeed to the 300-500GHz range, if recent news is accurate21:20
fennthat article (parallelism is not concurrency) is not making sense to me21:20
augurshame. i dont think i could explain it better than he does21:20
fenn"parallelism is an abstraction, not an implementation" huh?21:20
fenni think he should start off with some definitions after "rebooting the reader"21:21
auguralso, fenn, if megahal is a markov model, i can tell you /precisely/ why it sucs21:22
fennbecause it's stateless21:22
augurwell besides that21:22
augurthere are many reasons why megahal sucks21:22
fenni think it's quite fun actually21:22
augurthe first, which statelessness relates to, is that natural language is not regular, probably not even context-free. markov models are strictly regular.21:23
fenntell me why it sucks more than any other simple AI algorithm21:23
augurso its no surprise it produes garbled crap.21:23
fennyou never answered what the intention was21:23
augursecondly, its not really AI: its just chewing through text and running correlations to spit out new text, and would work similarly on, say, chess moves or musical notation or digestive patterns. it has no recourse to any sort of semantics to structure its behavior in any fashion.21:24
auguri did tell you my intention!21:24
augursomething agent-ish!21:25
fenn"its just chewing through text and running correlations to spit out new text"  is not intelligence?21:25
augurwell, you know, in as much as anything is intelligent :)21:25
fenni believe most humans do the same sort of thing, but have better filtering21:26
auguri wouldnt consider just running statistical analyses of symbol transitions by itself to be enough for something to be considered AI21:26
augurhumans definitely Do do similar sorts of things21:26
augurbut its almost certainly not _all_ we do21:26
fennn-gram transitions perhaps21:26
fennagain, this is getting lost in a text-based view of reality, which it certainly is not21:27
auguri mean, at some level maybe its transitions of some sort. some crazy higher-order transition stuff who knows21:27
augurthats also another problem -- megahal is just a text widget.21:27
fennand you expect your AI to make sense of non-textual data?21:28
fennsuddenly it's not a simple problem21:28
auguri dont expect it to make sense of anything significant. like i said, its intended to be an agent, so its non-textual data is just its affordances, its in-computer environment, etc.21:29
kanzurewheee isn't this fun21:31
kanzurealso, i told you so21:31
fennkanzure: wah.21:31
kanzureit's okay you have a chance to redeem yourself in robot hell21:32
fennaugur: you should do something with music, real-time sound analysis and repetition. lots of room for creativity21:32
kanzurewhat's wrong with whole brain emulation?21:32
fennit's too slow21:32
kanzurein the scheme of things that's not a big problem21:33
augurkanzure: nothing. its just not my goal right now. :)21:33
kanzurei thought your goal was 'ai'21:33
kanzure(and i thought my goal was goalism or something)21:33
kanzuregoal goal goal goal21:33
fennkanzure: would you like to play kurzweil for a day? when will we have sufficient computing power to simulate a human brain in real time for $1000?21:33
auguralso, i want to understand thought, not merely recreate it somehow21:33
augurblack boxes do not satisfy me21:33
kanzurefenn: umm, uhh, doesn't theuncertainfuture.com do that21:33
fennyeah but your specific prediction21:34
kanzurein fact, i'm pretty sure there's a few scenarios on there specifically for wbe21:34
augurfenn: i have some minor interest in music stuff21:34
kanzurewell every time i put numbers into their it says a singularity is happening in 2 or 3 years21:34
augurbut not anything related to analysis, etc.21:34
augurim more interested in formal theories21:34
fennformal theories are all bullshit21:34
kanzuremy mom was a formal theory21:34
kanzurei can confirm this21:34
kanzureformer formal theory21:34
fenn<- scientist, discordian21:35
fennaugur: there's not enough to work with if you just feed your baby curated text21:35
kanzuredoesn't ENKI do that21:35
fennuh, too many enki's21:36
kanzureENKI-][:21:36
kanzurethat one.21:36
fennresolve name conflict please!21:36
kanzuremegahal namcub accela guy21:36
kanzurejohn ohno21:36
fennuntil a bot shows up here and has a nice conversation here with me, i'll continue persisting in my beliefs about what will and will not work21:37
augurfenn: uh, my intention _isnt_ to feed it curated text of any real sort. not for training.21:37
fennwhat then?21:37
auguralso, formal theories arent all bullshit. they work plenty well.21:38
fennthe "if p then q" type don't work21:38
fennthe "this looks like a p, maybe q" stand a chance21:38
augurthats (classical) logic.21:38
augurthe majority of formal theories are  /not/ classical logic.21:39
fenndo enlighten21:39
kanzurepron.com user db? wtf http://lulzsecurity.com/releases/pronz.txt21:39
augurwell, when it comes to AI i cant say, actually21:39
kanzuresorry but this still doesn't sound productive at all21:40
augurbut for logic itself, there are plenty of non-classical logics. linear logic and its descendants are pretty interesting. one is a logic of "causality" that has some interesting capacities.21:40
augurframe-based reasoning works really well at text comprehension, at least from what ive seen.21:40
-!- eridu [~eridu@gateway/tor-sasl/eridu] has joined ##hplusroadmap21:41
augurand that was back in the 70s21:41
augurhey eridu.21:41
eriduhello augur21:41
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has joined ##hplusroadmap21:42
augurhows life, dude21:42
eridubusy as fuck21:42
fennthe wp article on frame reasoning mentions "spreading activation" which sounds a lot like neural networks to me21:43
fennare you saying neural networks are a formal theory?21:43
augureridu: anything good?21:44
eridunot really21:44
eridulot of half-finished projects and lost campaigns21:44
eriduI guess I got published in between now and when we last talked21:45
augurfenn: spreading activation in the frame context just means that there is a semantic network that structures the frames, whereby the frame problem is solved by biased search to the activated frames21:45
augurfenn: eg frame 1 "activates" frame 2, which means when the program looks for new frames to apply to the situation, itll check frame 2 first21:46
augurthats a toy explanation of spreading activation but you get the point21:46
augurthe frames are just formal objects akin to predicates with more information21:46
augureridu: oh? published for?21:46
augureridu: also, lost campaigns? :\21:47
augurdid you run for emperor?21:47
fennwhat creates the bias for frame 2 instead of some other frame?21:47
kanzureframe.. context.. words, bitches21:47
fennalso my google fu is failing at "logic of causality"21:47
eriduaugur: yeah, the administration fucked us over on daycon21:48
eridualso sds is dead21:48
fennhttp://en.wikipedia.org/wiki/Causality#Logic  looks like "if p then q" logic to me21:48
augurfenn: the background semantic network that the designers built in. things like restaurant frames are associated with eating frames, paying frames, etc. but not with, say, sky diving frames21:49
augurfenn: you'd think so, but thats because natural language is tricking you ;)21:49
augurfenn: classical implication (if-then) is strictly a correlational property, not a causational one.21:49
-!- eridu [~eridu@gateway/tor-sasl/eridu] has quit [Quit: Leaving]21:49
auguractually natural language's conditional is also correlational, but there's a strong pragmatic implicature in certain contexts21:50
augurthere are many ways to override this tho.21:50
auguri can find you the causal logic thing. they address AI-related issues in it, actually.21:50
augurclassical ones tho, nothing magical.21:50
augurProlegomena of a Logic of Causality and Dynamism - Bellot et al.21:51
kanzurei can never really size up people when talking about ai21:51
kanzurei have no idea if you're a machine learning newbie or not21:51
kanzurebut i do remember you wrote some linguistics papers?21:51
augurim a linguist, yes21:51
augurand im familiar with many ideas form ML, i just dont know jack about the details21:51
augurfenn: http://wellnowwhat.net/transfers/Prolegomena%20of%20a%20Logic%20of%20Causality%20and%20Dynamism%20-%20Bellot%20et%20al..pdf21:52
fennML is just a fancy word for "statistics with computers"21:53
kanzuresearch algorithms are statistics now?21:54
fennmost people think of statistics as what they learned in high school, so it really does need to be sexed up21:54
fennyeap pretty much21:54
kanzurethe mlai world just keeps redefining shit in a million ways21:55
kanzurethis is really useless guys..21:55
augurkanzure: ML is more than just search algos21:55
augurwell, depends on what you mean by search i guess21:55
kanzureyes i know21:55
augurultimately, fenn, ofcourse statistical inferences need to be used21:56
augurbut you cant do statistics without a good underlying model that is non-statistical21:57
fennyes we need to implement it after all21:57
fennhave you ever read "reciprocality"?21:58
augurnope21:58
-!- mayko [~mayko@71-22-217-151.gar.clearwire-wmx.net] has quit [Ping timeout: 240 seconds]21:58
augurstatistics is an addition to an underlying formal theory tho, i must stress this21:58
fennhttp://www.buildfreedom.com/content/reciprocality/21:58
fennformal theories are what you do with the last step, once you know it's probably a cat and not a chair21:59
augurnot at all true21:59
fennthe problem seems to be determining it's a cat21:59
auguryou simply cant do any statistics without an underlying non-statistical theory21:59
augursimply because without that, theres nothing to run statistics over!21:59
fennyeah sure, like "this is a matrix with pixel values from a camera"21:59
fennthat's not a theory, that's a statement of fact22:00
auguryeah but what statistics will you calculate off of that22:00
fenni'd run lots of things in parallel22:00
auguryeah but what statistics22:00
fenncorrelation between feature clusters22:00
augurfeature clusters are not in the pixel values22:01
fennright22:01
augurcongratulations, youve just invented a theory22:01
augurand a formal one at that22:01
fenni am underwhelmed22:01
augurbecause the feature clusters are below the stats!22:01
auguryou should be underwhelmed!22:01
augurvision theory is a different issue tho22:02
augurofcourse AI should have a statistical vision theory underlying it22:02
fenn(i mean feature as in the SIFT points, not in the statistics sense)22:02
augurbut vision theory itself, as a whole, has the stats on the top.22:02
augurif you stratify to separate concerns, you get the formal theory at the bottom of each module, with stats on top of that22:03
augurstats over a theory of the visual system, stats over a theory of sentence structure, stats over a theory of world knowledge, stats over a theory of human behavior, etc.22:03
fenni don't like hard coding theories22:04
auguryou have to have _something_ to calculate stats over. even if its just your theory of visual features.22:04
fenncurrently with limited computer power can only perform brittle crystalline calculations on pixel values, but it doesn't *have* to be that way22:05
auguror you theory of theories of visual features, if you want to run higher-order statistics22:05
auguri agree completely.22:05
augurbut you still have to have a theory of what matters and what doesnt, what is looked for and what isnt22:05
fennif you say so22:05
augurthe intuitive stuff like edges, polygons, shading, etc. are all sensible, ofcourse22:06
augurbut those arent derived from the stats22:06
augurthey precede it22:06
fenn"of course" is what killed AI research22:06
augur;)22:06
epitronnature doesn't have many polygons :(22:06
fenn"of course space is euclidean"22:06
augurepitron: sure it does, they're just really tiny ;)22:07
epitronand in terms of how the brain works, polygons are the highest level22:07
augurfenn: well, afaact it IS euclidean!22:07
epitronthey're very abstract and generalized22:07
epitronlike language22:07
epitronyour modules are upside-down22:07
epitronyou're doing top-down plus bottom up statistics22:07
augurepitron: im not saying polys really are at the bottom22:07
auguri was using that as an example22:08
epitronsure22:08
epitronbut it's still upside-down :)22:08
augurperhaps!22:08
augurmy point is that you cant run statistics over nothing22:08
fennthe egyptians drew their maps with the source of the nile at the top22:09
auguryou always have to make decisions about what the underlying theory is.22:09
augurand that underlying theory is not itself a statistical one22:09
epitronyou mean, assumptions to limit your search space?22:09
epitronthose are good22:09
fennhe just means how you determine the set of data to do statistics upon22:09
epitronwell, if they're good assumptions22:09
epitrontrue22:09
epitrongood data is lso good :)22:10
fennit's not even a search space at that point22:10
augurfenn: often enough the data you run stats on is structured in some way or other22:10
fennlet's just say it's binary data22:10
fennwhat regularities do you see?22:10
epitroni wonder how many people in the world are talking about re-engineering the brain right now22:10
epitronprobably a lot22:10
augurfenn: that depends on what sort of grammar you're using!22:10
epitronat least a few hundred22:10
fenncan you infer ascii text based on the absence of a 1 in the 8th bit?22:10
augurif you're using a 1-gram model, you get jack shit22:10
augurif you're using a 10-gram model you get more22:11
augurif you allow context free behavior, you get even more22:11
* epitron rolls a 1-gram model22:11
fenna few hundred is not a lot compared to most fields22:12
augurfenn: you have to make a choice even in binary processing about what sort of structure the data is assumed to have22:12
augurif you choose a CF model, you will get better results than if you have a regular model. that is a fact.22:13
augurat least if the data is CF in structure22:13
fennCF?22:13
augurcontext-free22:13
fennif you start with the answer, you will get a better answer22:14
augura classical example of a CF language, in pure binary, would be: all bit strings with a bunch of 0's at the beginning, and the same number of 1's folloiwng that22:14
augura markov model simply can never learn this. it might _MIGHT_ be able to get arbitrarily close. but that iwll also depend on how many states you allow it to have, etc. etc.22:15
augurfenn: im not starting with the answer tho thats the point22:15
kanzure'america invents act' http://www.hoover.org/publications/defining-ideas/article/8209622:15
auguryoure always making assumptions about the nature of the system. assumptions which arent based on statistics.22:15
augursure you can generalize your system so that it can dynamically shift from one kind of assumption to the next, or whatever, but you're not escaping the underlying non-statistical nature of the system22:16
fennyou can start with a large number of statistical models and choose the one that predicts the data best22:16
augurall you're doing is making the non-statistical component adaptive22:17
augurfenn: sure, but those statistical models are still statistics over SOMETHING22:17
augurand that something is not statistical in nature22:17
fennnoisy sensor data22:17
augurnot just22:17
fenndon't get all "digital sampling is a formal theory" on me22:17
auguryour stats are stats of SOMETHING22:18
augurstate transitions, production rules, etc.22:18
augurthey're stats running over something non-statistical22:18
fennis a fourier transform a formal theory?22:18
augurthat depends on your opinion of math. it is certainly a purely mathematical concept.22:19
augurbut its also tangential to the point22:19
auguryou run a fourier transform, but /then what/?22:19
augurnow you just have data of a different shape.22:19
augurbig deal.22:19
augurwhat are you attributing probabilities to, fenn22:20
augurprobabilites are probabilities of _something_22:20
fennit is an assumption to make in order to get better predictability out of your statistics22:20
augursure, fourier transforms are nicer to deal with. i know that22:20
augurbut that doesnt change anything22:20
augurwhat are you calculating the probabilities of. thats the point.22:20
augurwhat has the probabilities.22:20
fennarbitrary things22:21
fenni dont know the word22:21
fenna variable22:21
auguri feel like you're not very familiar with machine learning...22:21
fennmonica calls them 'hypotheses'22:21
augurwell i dont know what monica is22:22
fenna person who does ai research22:22
auguryyyeah...22:22
fenndifferent algorithms have different sorts of outputs, it's not always straightforward to integrate many different algorithms22:23
kanzureare you two going to blather about bayesian bullshit yet22:23
kanzure*poke*22:23
augurlook, its clear you're not familiar enough with ML techniques to even defend your own position adequately22:23
* kanzure pops some popcorn22:23
QuantumGsooo.. linguistics eh.. are dictionaries descriptive or prescriptive?22:23
augurkanzure: well, bayesian inference is ofcourse superior, so.22:23
augurQuantumG: depends on the dictionary.22:24
QuantumGcan a prescriptive dictionary be the result of linguistics?22:24
augurno, but i would argue that descriptive ones cant either.22:24
fennif it's a conlang22:24
QuantumGdo tell22:25
augurlinguistics is the scientific study of language22:25
augurthat immediately rules out prescriptivism in any form22:25
QuantumGagreed22:25
augurscience describes, models, etc. it does not proscribe22:25
QuantumGgo on22:26
augurfurther, dictionaries are documentations of usage. documentation is important, but its not study. there is no explanation to be done with dictionaries.22:26
QuantumGahh.. I see your argument.  Well, intermediate products are also products.22:27
augurthey are, but they're not necessarily the same sort of thing as the end product.22:27
augurYou need gasoline for cars, but oil refinement is not automotive engineering.22:28
auguri have great respect for lexicographers, but its not linguistics.22:28
augurthere is no science being done there.22:28
fenni dont see how you can do science by making formal models of a system that routinely breaks the rules22:29
auguryou'd be surprised how little language actually violates the rules.22:30
auguryou just have to know what the rules actually are.22:30
QuantumGdiscovering that there are any rules at all is the greatest achievement of linguistics22:30
fennmeh22:30
augurmeh!22:31
fenngrammar goes back thousands of years22:31
auguryes it does, but not as a science22:31
augurgrammar as a prescriptive act certainly does tho.22:31
QuantumGphysics goes back thousands of years.. science is the method of refinement22:31
fennwhich came first, the language or the prescriptivist grammar?22:31
augurthe language, surely.22:32
auguryou cant be snobby about french if there aint no french!22:32
fennso where did the grammar come from?22:32
augurgrammar is in peoples heads.22:32
QuantumGphysical reality of course.22:32
augurwriting down a grammar for a language is just an attempt to capture the knowledge people have in their heads22:33
QuantumGlinguistics is as much about the study of physical reality as physics.22:33
fennsome people would say genetics or evolution, but i'm not sure22:33
augurfenn: thats where the FACULTY of language comes from22:33
fennoh the FACULTY22:33
augurbut the grammars themselves are not genetic/evolved22:33
fennthat makes it so clear!22:33
augurfenn: think of it like this22:33
augurbeing able to move around on two legs coordinatedly is genetic22:34
augurwaltzes are not.22:34
fennferal children raised by wolves do not walk on two legs22:34
augurthis is (perhaps?) true22:34
augurthe brain is not static22:34
fennalso there are dogs with only two hind legs that balance22:34
augurunused portions of the brain are reallocated.22:34
augurwe know this.22:35
fennif all humans walked on four legs, so would their children22:35
augurprobably!22:35
fennwhat makes waltzes less genetic than walking?22:35
augurbut take one of their children and raise the kid among bipeds, and the child would not have trouble learning it competently22:35
augurand without any training, too22:36
augurbut a dog will not.22:36
fennlet us imagine a machine that can perform any possible sequence of foot maneuvers22:36
auguradmittedly they probably have the cognitive capacity, its mostly in their physical structure thats an issue22:36
augurthey also probably lack the finer balance stuff that bipeds have22:36
fenni bet you a bajillion dollars most humans will not be able to learn a significant portion of its dances22:36
augurcertainly22:36
augurits an analogy, fenn22:37
QuantumGnot that it's the topic, but I don't think we'd consider them "dances" .. just as we don't consider many sequences of sounds to be language.22:37
augurpoint is, languages are not genetically encoded. the ability to use them is.22:37
fenni disagree22:37
augurwell you're wrong.22:37
fennit has been a displeasure talking with you22:38
augurthe feeling is mutual, o lay friend.22:38
* kanzure eats more popcorn22:38
kanzurewhen will you mortals learn that i am the superior option22:38
fennwhat's the inverse of "told you so"?22:39
kanzure"you sure told me"?22:39
augurfenn: "you're right"22:39
auguror "you were right"22:39
fennbtw, space is not euclidean, in case that was in doubt.22:53
augurall measurements show it to be flat, fenn.22:54
augurthere's no detectable curvature22:54
auguryou know, modulo gravity22:54
QuantumGall evidence indications that space is euclidean22:55
augurthe large-scale geometry of space is flat. locally, gravity changes things ofcourse.22:55
* augur steals kanzure's popcorn23:00
-!- mode/##hplusroadmap [+o kanzure] by ChanServ23:02
-!- augur was kicked from ##hplusroadmap by kanzure [augur]23:02
@kanzuredon't fucking steal my motherfucking popcorn23:02
QuantumGheh23:02
-!- mode/##hplusroadmap [-o kanzure] by ChanServ23:02
-!- augur [~augur@208.58.6.161] has joined ##hplusroadmap23:03
augur:|23:03
* augur knocks kanzure's popcorn out of his hands23:03
augur>|23:03
fennreally, there arent any articles written about "culture invariance of grammar"?23:08
augursorry?23:09
augurwhat are you looking for, i can direct you23:10
fennwhat is the verbal equivalent of phosphenes called?23:10
augurwhat23:10
fenncharacteristics of grammar that are preserved across all cultures23:10
QuantumGculture invariance of grammar = universal grammar23:10
QuantumGthere's an absurd number of articles about it.23:10
auguri dont think that phosphenes are what you mean23:10
augurbut yes, grammatical universals are the property in question23:11
auguruniversal grammar is the theory of grammatical universals23:11
augurmaybe you're thinking of form constants, fenn?23:12
QuantumGstudy of how children "flip the bits" on the universal grammar to learn the particular native dialect is about a year's worth of reading.23:12
augurhonestly, im skeptical of that sort of parametricity, QuantumG.23:12
augurive never bought it, to be honest.23:12
fennno, not form constants23:12
fennphosphenes are patterns and colors you see when you push on your eyeballs23:13
fennall humans see the same patterns23:13
augurperhaps23:13
augurif so, only because of the structure of the eye23:14
augurform constants are a better analogy to universals.23:14
-!- jenzebubble [jen@173-29-174-114.client.mchsi.com] has quit [Ping timeout: 258 seconds]23:14
QuantumGwhich is the same argument for universal grammar.  That we all have the same machinery for "looking" at language.23:14
auguractually if there were an eye-ball analogy it would relate to ears :p23:15
QuantumGwell, I was more thinking about edge detection and other visual cortex processing that we all do.23:16
fennthe drawings online of form constants dont look anything like phosphene shapes to me23:16
QuantumGbut yes23:16
augurfenn: no, they dont23:16
augurthey're not supposed to23:16
augurform constants are shapes that are common to many sorts of atypical visual stimulation23:16
fennoften phosphenes look like characters written in some language i can't read, or grid or moire patterns23:17
augureg hallucinogens23:17
auguryou have interesting phosphenes23:18
augurive never had that23:18
fennwell, so much for that theory :P23:19
augurwhat23:22
fennthat all humans see the same patterns23:22
fennprobably just me misremembering what i read23:23
auguryou read about form constants23:23
mrtrousersAnyone interested in coming to Barcelona and get a Hackerspace started?23:23
mrtrousersThere will be biohacking!23:23
augurmrtrousers: i wish i could!23:23
mrtrousers;)23:29
-!- klord [~klord@99-67-239-104.lightspeed.austtx.sbcglobal.net] has quit [Ping timeout: 252 seconds]23:49
-!- klord [~klord@99-67-239-104.lightspeed.austtx.sbcglobal.net] has joined ##hplusroadmap23:57
--- Log closed Tue Jun 14 00:00:07 2011

Generated by irclog2html.py 2.15.0.dev0 by Marius Gedminas - find it at mg.pov.lt!